real-time imaging, forecasting, and management of human …gljpv/pdfs/clarke_etal_2019_srl.pdf ·...

Real-Time Imaging, Forecasting, andManagement of Human-Induced Seismicity atPreston New Road, Lancashire, Englandby Huw Clarke, James P. Verdon, Tom Kettlety, Alan F. Baird,and J-Michael Kendall

ABSTRACT

Earthquakes induced by subsurface fluid injection pose a signifi-cant issue across a range of industries. Debate continues as to themost effective methods to mitigate the resulting seismic hazard.Observations of induced seismicity indicate that the rate of seis-micity scales with the injection volume and that events followthe Gutenberg–Richter distribution. These two inferences per-mit us to populate statistical models of the seismicity andextrapolate them to make forecasts of the expected event mag-nitudes as injection continues. Here, we describe a shale gas sitewhere this approach was used in real time to make operationaldecisions during hydraulic fracturing operations.Microseismic observations revealed the intersection betweenhydraulic fracturing and a pre-existing fault or fracturenetwork that became seismically active. Although “red light”events, requiring a pause to the injection program, occurredon several occasions, the observed event magnitudes fell withinexpected levels based on the extrapolated statistical models, andthe levels of seismicity remained within acceptable limits asdefined by the regulator. To date, induced seismicity has typicallybeen regulated using retroactive traffic light schemes. This studyshows that the use of high-quality microseismic observations topopulate statistical models that forecast expected event magni-tudes can provide a more effective approach.

INTRODUCTION

Human-induced seismicity is becoming an increasingly contro-versial topic. It is well known that activities such as mining andwater impoundment can lead to felt seismicity, but increasingly,activities such as geothermal energy (Grigoli et al., 2018),underground storage of waste such as CO2 or water(Keranen et al., 2014), production from conventional hydro-carbon reservoirs (e.g., Segall, 1989) and hydraulic stimulationof shale gas reservoirs (Bao and Eaton, 2016) are attractingconcern from the public, regulators, and operators.

The stimulation of fractures by injecting water at highpressure is a technique used to create conductive fracture net-works in low-permeability reservoir rocks. Hydraulic fracture

stimulation is widely used in the commercial production ofhydrocarbons and in developing engineered geothermal sys-tems. Use of this method has become more prominent inthe past decade, associated primarily with the shale gas boom(Wang and Krupnick, 2013) in North America.

If hydraulic fractures intersect a pre-existing fault that isnear its critical stress state, the increase in pore pressure canreduce the effective normal stress, declamping the fault andcreating induced seismicity. Such cases are relatively rare:Atkinson et al. (2016) estimate that only 0.3% of wells inBritish Columbia and Alberta, a region with some of the high-est levels of hydraulic fracturing-induced seismicity (HF-IS),are associated with induced events larger than magnitude 3.Nonetheless, the issue of induced seismicity is a concern forthe petroleum and geothermal industries and will likely beof concern to other nascent industries, such as carbon captureand storage, as well (e.g., Verdon, 2014).

Debate continues with regards to the most effective meth-ods to mitigate HF-IS and what regulations should be applied.To date, regulators have typically imposed traffic light schemes(TLSs) whereby the operator reduces, pauses, or stops injectionif the magnitude of the largest event exceeds a specified thresh-old. TLS thresholds have varied significantly in different juris-dictions (Bosman et al., 2016; Baisch et al., 2019). For example,whereas in Alberta, the red light is set atM � 4, in the UnitedKingdom, the red light is set atM � 0:5, a difference in earth-quake moment of more than 175,000 times.

The simple TLSs currently used by hydraulic fracturingregulators are essentially retroactive in nature because the oper-ator takes actions after an event has occurred. In some casestudies, seismicity has been observed to continue and toincrease in magnitude after injection has ceased (e.g.,Häring et al., 2008; Clarke et al., 2014). These postinjectionincreased-magnitude events, known as “trailing events,” posean issue for TLSs because they compel the regulator to setthresholds that may be substantially lower than the actual mag-nitude they wish to avoid. Hence, operations may be stoppedeven though levels of seismicity are well below that whichmight be considered hazardous.

1902 Seismological Research Letters Volume 90, Number 5 September/October 2019 doi: 10.1785/0220190110

Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/90/5/1902/4825110/srl-2019110.1.pdfby JamesVerdon on 06 September 2019

It is therefore desirable to manage and mitigate inducedseismicity in real time, as operations proceed. For example,injection volumes or pressures could be reduced (e.g., Kwiateket al., 2019), or stimulation can be directed away from areasshowing fault reactivation. Here, we show a successful exampleof managing HF-IS with a recently acquired dataset from ashale gas operation in the United Kingdom.

Using Microseismic Data for Decision Making toMitigate Induced SeismicityThe TLSs described by Bosman et al. (2016) and Baisch et al.(2019) that are currently used to regulate hydraulic fracturingstipulate decisions based solely on the magnitude of the largestevents. This is a rational option if monitoring is provided bynational or regional seismometer networks, where monitoringstations may be 10s of kilometers from the site (e.g., Clarkeet al., 2014; Friberg et al., 2014; Schultz et al., 2015; Skoumalet al., 2015). In such cases, only the larger events may bedetected, and hypocentral locations and focal mechanismsmay be poorly constrained. Hence, the only reliable, well-con-strained data are the magnitudes of the larger events.

However, it is common for operators to deploy microseis-mic monitoring, in which downhole geophone arrays (Maxwellet al., 2010) or dense surface arrays (Chambers et al., 2010) areable to detect very low-magnitude “microseismic” events.High-quality microseismic monitoring may record thousandsor even hundreds of thousands of events with very precise loca-tions, spanning several orders of magnitude, provided in realtime during operations (e.g., Zinno et al., 1998). These datawill be highly relevant for understanding the risks posed byHF-IS. However, such data are not used by the relatively simpleTLSs currently being applied by hydraulic fracturing regulators(Bosman et al., 2016).

There are two primary ways by which microseismic obser-vations can be used to guide decisions to mitigate induced seis-micity. First, microseismic data can be used to detect andcharacterize the interactions between hydraulic fractures andpre-existing faults (Maxwell et al., 2008, 2009; Wessels et al.,2011; Eyre et al., 2019; Igonin et al., 2019; Kettlety et al.,2019). Microseismic events during hydraulic fracturing typi-cally occur in clusters extending from the well perpendicular tothe minimum horizontal stress, tracking the growth of thehydraulic fractures and mapping the extent of the stimulatedreservoir volume. If a fault is intersected, events may begin toline up along the structure, allowing it to be identified andmapped (e.g., Maxwell et al., 2008; Wessels et al., 2011;Hammack et al., 2014; Eyre et al., 2019; Igonin et al.,2019; Kettlety et al., 2019). In many cases, fault reactivationcan also be identified by a decrease in Gutenberg and Richter(1944; hereafter, G-R) b-values (e.g., Maxwell et al., 2009;Verdon and Budge, 2018; Kettlety et al., 2019) or by anincrease in the rate of microseismicity relative to the injectionrate (e.g., Maxwell et al., 2008; Verdon and Budge, 2018).

If a fault is identified during injection, then an operatorcan redesign the injection program to avoid further interactingwith the fault. This can be achieved, for example, by skipping

stages along a horizontal well, changing the planned injectionrates or volumes, or altering the properties of the injected fluid(e.g., a more viscous fluid will carry more proppant while trav-eling less distance into the formation). Alternatively, Hofmannet al. (2018) proposed adopting a “cyclic soft stimulation” pro-gram in which repeated injection is conducted at significantlylower rates. Zang et al. (2019) demonstrated this approach forexperimental-scale injection tests. However, the results fromapplication to an industrial-scale project (Hofmann et al.,2019) are more ambiguous because the Pohang geothermalproject, South Korea, at which this method was applied, wenton to experience one of the largest injection-induced eventsever recorded (Grigoli et al., 2018). Moreover, for shale gashydraulic fracturing applications, it is not clear that such alow-rate injection program would result in effective proppantplacement into a shale formation.

Microseismic data can also be used to make forecasts of theexpected event magnitudes during stimulation. Induced seismic-ity has been observed to follow the G-R distribution (van derElst et al., 2016), with the total number of events (Shapiro et al.,2010; Mignan et al., 2017) or the cumulative seismic momentreleased (Hallo et al., 2014) being scaled to the cumulative injec-tion volume. As such, expected event magnitudes can be forecastby characterizing these relationships for the site in question andthen extrapolating them to the planned injection volume. Thisapproach has shown significant promise when applied in apseudoprospective manner (e.g., Verdon and Budge, 2018).

These concepts have produced more advanced approachesto mitigate induced seismicity. For example, Mignan et al.,(2017) propose an adaptive traffic light scheme (ATLS),whereby the daily rate of seismicity is scaled to the injectionrate (as per Shapiro et al., 2010), with the addition of a post-injection relaxation time that describes trailing effects. Eventmagnitudes are then determined from a G-R distribution, fromwhich risk-based decisions can be made. Broccardo et al.(2017) extended the Mignan et al. (2017) approach by provid-ing a Bayesian framework within which the key parameters canbe estimated. However, to our knowledge, this approach hasnot yet been applied in real time to an active project.

Kwiatek et al. (2019) present an example of such methodsbeing applied in real time to a deep geothermal project nearHelsinki, Finland. They found that the observed seismicityscaled with injection parameters, allowing them to adjustthe injection program to ensure that the levels of seismicityremained within the limits imposed by the regulator. The suc-cess of the type of approach demonstrated by Kwiatek et al.(2019) and the continued refinement of proposed adaptiveTLSs (e.g., Broccardo et al., 2017; Mignan et al., 2017) providethe opportunity to move beyond the simple TLSs currently incommon usage. However, their effectiveness must be demon-strated extensively in real-time scenarios so that regulators gainconfidence in their application.

A Case Study from Northwest EnglandIn this article, we report on the Preston New Road-1z(PNR-1z) well, Lancashire, United Kingdom, operated by

Seismological Research Letters Volume 90, Number 5 September/October 2019 1903


Cuadrilla Resources Ltd. (CRL). This was thefirst United Kingdom onshore well to behydraulically fractured since a governmentreview of HF-IS seismicity was concluded in2012. As such, it was the subject of regularnational media attention (e.g., Webster, 2018)and debate in the national parliament(Hansard, 2018). Given the high levels of publicscrutiny, the site was extensively monitoredboth by CRL and by independently fundedorganizations such as the British GeologicalSurvey (BGS). This monitoring includedgroundwater, surface water, air quality, and traf-fic movements, as well as the induced seismicitymonitoring described here. Extensive baselinesurveys were conducted for all of these items,so that any change from the pre-operationalconditions could be identified.

Given public concerns about HF-IS in theUnited Kingdom, CRL took proactive measuresto mitigate induced seismicity, guided by micro-seismic observations as outlined previously.Here, we provide a brief description of the oper-ations conducted at the site and then show howmicroseismic data were used to identify andmap the interaction between hydraulic fracturesand a fault and to forecast expected event magnitudes as theinjection progressed. This information allowed CRL to adjustits injection program, ensuring that levels of seismicity did notexceed the overall objectives set by the regulator, as well as pro-viding an increased understanding of more proactive measuresthat could be applied in future as alternatives to simplis-tic TLSs.

DESCRIPTION OF THE PNR-1Z SITE

The PNR-1z well targets the Carboniferous Bowland Shale at adepth of ∼2300 m. The lateral portion of the well extends780 m in a westward direction (Fig. 1). A sliding-sleeve com-pletion was used, with 41 individual sections spaced at 17.5 mintervals. CRL planned to stimulate each of these sleeves with400 m3 of slickwater, placing 50 tons of proppant per sleeve.Stimulation was carried out in two periods (Fig. 2), first from15 October to 2 November and then from 8 to 17December 2018.

United Kingdom Regulations for Induced SeismicityIn the United Kingdom, HF-IS is regulated by the Oil and GasAuthority (OGA). The OGA’s objectives are to minimize thenumber of events felt at the surface by the public and to avoidthe possibility of events capable of causing damage to nearbybuildings or infrastructure (Oil and Gas Authority, 2018).United Kingdom standards for ground vibrations from otheractivities such as quarry blasting, construction equipment, andindustrial machinery are provided by British Standard BS7385-2. This sets a peak ground velocity threshold, above

which cosmetic damage, such as cracking of plaster, of15 mm=s (at lower frequencies such as would be expected frominduced seismicity), may result. Using ground-motion predic-tion equations (Akkar et al., 2014) for hypocentral depthsequivalent to expected depths of hydraulic fracturing andmaking conservative assumptions for ground conditions, thisthreshold is approximately equivalent to a magnitude ofM � 3:5. Therefore, the OGA’s objective could be reasonablytranslated as minimizing the number of events that havemagnitudes 2 < M < 3 and avoiding events that have magni-tudes M > 3:5.

To regulate HF-IS, the OGA currently applies aTLS witha red-light threshold of M � 0:5 (Green et al., 2012), forwhich the operator must stop injection, reduce the pressure inthe well, perform well integrity checks, and wait at least 18 hrbefore resuming injection. This is by some margin the moststringent level for ground motion applied to any industrialactivity that we are aware of. TheM � 0:5 red-light thresholdis 175 times smaller than the M � 2 events that the schemeseeks to minimize and more than 30,000 times smaller than theM > 3:5 events that the scheme seeks to avoid. This disparityexists to mitigate the risk posed by trailing events, in whichevent magnitudes may continue to increase after injectionhas been stopped (see Mignan et al., 2017, for an attemptto forecast trailing event populations). This TLS was appliedto stimulation of the PNR-1z well, and the restrictive nature ofthis scheme had a significant impact on the operations: only 17of the planned 41 stages were injected, and of these, only twoinjected the 50 tons of proppant that was planned. However,only two events were reported by the BGS as being felt, and

▴ Figure 1. Map of operations at Preston New Road (PNR) showing the positionsof the drilling pad and horizontal tracks of PNR-1z and PNR-2 and the positions ofthe surface monitoring stations. The black box marks the area of interest shown insubsequent figures. Major roads, rail links, and nearby villages are also marked.Coordinates are United Kingdom (UK) Grid Reference.

1904 Seismological Research Letters Volume 90, Number 5 September/October 2019


ground motions remained well below the levels at whichdamage might be expected. Therefore, overall, the operationcomplied with the regulator’s objective to minimize feltseismicity and avoid damaging seismicity.

Real-Time Seismic and Microseismic MonitoringTwo systems were used in combination to monitor inducedseismicity at PNR. Both of these systems provided event loca-tions and magnitudes in real time (typically within 1–4 min ofevent occurrence) computed by a processing contractor(Schlumberger). To administer the TLS, an array of eight sen-sors, including two broadband seismometers and six geophones(4.5 Hz instruments), was deployed at the surface, augmentedby four broadband seismometers deployed by the BGS (Fig. 1).During real-time monitoring, the surface array identified 54events with a minimum magnitude of ML � −0:8. Thesurface array provided sufficient coverage such that focal

mechanisms could be determined for nine ofthe largest events during real-time monitoring.

Microseismicity was recorded using anarray of 24 geophones (15 Hz instruments)placed in the build section (where the well devi-ates from vertical to horizontal) of the adjacentPNR-2 well, 200 m shallower and 220 m north-east of the nearest sleeve in PNR-1z (Fig. 3).This array reported more than 39,000 eventsin real time, with a minimum magnitudeof Mw � −3:0.

A Note on MagnitudesMeasurements of magnitudes for small eventscan be challenging (Kendall et al., 2019). Twodifferent magnitude scales were in use duringreal-time operations at PNR. The UnitedKingdom TLS regulations mandate the use of alocal magnitude scale with a correction appliedto account for the small source–receiver distan-ces (Butcher et al., 2017; Luckett et al., 2019).Therefore, magnitudes from the surface arraywere reported asML values. However, theseMLscales are calibrated using surface stations,implicitly including free-surface effects andnear-surface attenuation, so thisML scale is notcalibrated for downhole instruments. Insteadthe downhole events were reported as Mwvalues. Although a direct comparison and con-version between the two scales might seem likean obvious solution (e.g., Edwards and Douglas,2014), this was more challenging in practice.The surface array recorded the largest 54 events,so only these events had reported ML values.However, many of these larger events producedsubsurface motions that were beyond thedynamic range of the downhole instruments, soaccurate downhole Mw values could not be

determined for these events. Hence, there is only a small subsetof events that are large enough such that a robustMw value canbe computed using the surface array but not so large such that arobust Mw value can also be computed using the downholestations, thereby enabling a comparison to be made.

Work is ongoing to resolve the observed ML and Mwvalues. However, the need for rapid decision making meantthat this information was not used during real-time operations.Instead, we usedML values for the 54 events that were reportedby the surface array and used Mw values for the remainingevents. Clearly, this solution was far from optimal. However,we note that doing so does not produce anomalies or unusualbehavior if the overall magnitude–frequency distribution isexamined (Fig. 4), suggesting that this approach was reasonablein this case. However, in future cases, this issue should beaddressed by ensuring that moment magnitudes are reportedby both array types and that relationships to convert betweendownhole Mw values and surface ML values are calibrated. In

▴ Figure 2. Overview of injection into PNR-1z. (a) The volume of fluid (blue) andmass of proppant (purple) injected into each sleeve. It also shows all M > 0 trafficlight scheme (TLS) events (yellow and red dots) that occurred during or after injec-tion into each sleeve. (b) Cumulative fluid volume (blue) and proppant mass (purple)injected as a function of time, again showing the occurrence of TLS events. Thenumbering in (b) shows the sleeve being injected. The background colors show theTLS green, amber, and red magnitude thresholds.



Figure 4, we fit a G-R distribution to the entire event catalogusing the Aki (1965) maximum-likelihood approach, comput-ing the magnitude of completeness, MMIN, using both theWiemer and Wyss (2000) formulation with an acceptancethreshold of 95%, which gave MMIN � −0:95, and by usinga Kolmogorov–Smirnov test with a 10% significance threshold(e.g., Clauset et al., 2009; Williams and Le Calvez, 2013),which gave MMIN � −0:8. In both cases, the resulting G-Rparameters were a � 1:9 and b � 1:3.

MICROSEISMIC OBSERVATIONS

Figure 5 shows a map and cross section for located events witha signal-to-noise ratio >5. Events during each stage are mostlyfound in the vicinity of the corresponding injection sleeve,extending ∼200 m to the north. The events extend ∼150 mabove and below the well, remaining within the Bowland ShaleFormation. The largest observed event has a magnitude ofM � 1:5, and in total, eight events exceeded theTLSM � 0:5threshold; whereas three of these occurred during injection andrequired pumping to be stopped, the remaining five were trail-ing events that occurred after injection had ceased.

Relationship between Microseismicity andPreviously Observed FaultsBefore the start of operations, a 3D reflectionseismic survey was acquired at the site.Several pre-existing faults and “seismic discon-tinuities” (i.e., potential small faults that are atthe limit of resolution for 3D seismic surveys)were identified (Cuadrilla Resources Ltd.,2018). We observed little or no correlationbetween the positions of these features andmicroseismicity. The events associated withstages 1–3 at the toe of the well overlap withone of the seismic discontinuities. However,the levels of microseismicity produced by thesestages were among the lowest. In contrast, noneof the events that exceeded the M > 0:5 TLSthreshold occurred on structures identifiedfrom the 3D survey.

Indeed, no microseismicity coincided withany of the large faults identified in the 3D seis-mic survey, all of which were significantly fur-ther from the well than the greatest distancesreached by the microseismicity. This observa-tion allowed CRL to proceed with confidencethat the hydraulic stimulation was unlikely tocause reactivation of the larger faults that hadbeen identified.

Identification of Potential SeismogenicStructuresThe northward propagation of microseismicityfrom each injection sleeve traces the propaga-tion of hydraulic fractures perpendicular to

the minimum horizontal stress azimuth of ∼80° (Fellgett et al.,2017). However, our interest was to identify pre-existing struc-tures on which the larger events may occur. We note that thelargest event, with a magnitude of M � 1:5, could correspondto a rupture with displacement of <1 cm with a length<100 m. At this scale, the distinction between a “small fault”and a “large fracture” is somewhat arbitrary: we will use “fault”hereafter to describe such features while keeping this factin mind.

In Figure 5, the events do not display an obvious align-ment along a pre-existing fault, an observation that often pro-vides the clearest evidence of fault reactivation (e.g., Eyre et al.,2019; Igonin et al., 2019; Kettlety et al., 2019). Instead, we usea combination of observations to identify and define the seis-mogenic structures responsible for the largest events.

Focal MechanismsThe focal mechanisms for six of the largest events are shown inFigure 6a. The events all have similar mechanisms: either left-lateral strike slip on a near-vertical fault striking northeast–southwest or right-lateral strike slip on a near-vertical faultstriking northwest–southeast. The consistent orientation of

▴ Figure 3. (a) Map and (b) cross section of the downhole monitoring arraydeployed in well PNR-2 and the sleeves through which injection was conductedin PNR-1z.



these focal mechanisms provides a constraint for the orienta-tion of any potential seismogenic structure.

Mapping Large Events and Cumulative Moment ReleaseFigure 6a also shows the positions of all events withM > 0 and maps the cumulative seismic moment release ΣM0.These observations allow us to identify a single zone in whichalmost all of the larger events were occurring and within whichthe overall cumulative seismic moment release was highest.This zone intersects the PNR-1z well at roughly the positionof Sleeve 18, which was the first stage on which an eventexceeding the M > 0:5 TLS threshold occurred. Interactionbetween injection activities and this zone occurred alongthe well toward the heel. Importantly, the orientation of thiszone matches the orientation of the northeast–southwest planeof the observed focal mechanisms.

Microseismicity during Injection HiatusThese observations allowed us to identify the seismogenic fea-ture during the initial stimulation of stages 18–41 in October2018 (Fig. 2). From 3 November, CRL paused the injectionprogram in response to repeated M > 0:5 events that hadoccurred during the previous week. The injection pause con-tinued until 7 December. Observations of microseismicity dur-ing this injection hiatus (Fig. 6b) provided the final anddefinitive identification of the seismogenic structure. Theevents during hiatus, almost all of which had magnitudes lessthan M < −1, were all located along the same feature that wehad identified from the focal mechanism orientations, the posi-tions of the largest events, and the cumulative momentrelease map.

Our overall interpretation of the observed microseismicityis that a pre-existing fault plane runs northeast from the well.During hydraulic stimulation, larger events occurred when thehydraulic fractures from each stage intersected this fault.During the hiatus, whereas the microseismic events associatedwith hydraulic fracturing stopped, low levels of microseismicitycontinued to persist along this feature for a longer period oftime. We fit a plane to a combined population of the M > 0events (Fig. 6a) and the hiatus events (Fig. 6b), by finding theplane that minimizes the least-squares distance between eachevent and the plane. We found a strike of 237° and a dip of70°, which are consistent with the observed focal mechanisms.We term this fault NEF-1 (northeast fault-1) hereafter. Withthe maximum and minimum horizontal stresses orientednorth–south and east–west, respectively, this plane is well-oriented for the observed left-lateral strike-slip motion, andthe observed focal mechanisms are therefore consistent withthe local stress conditions.

STATISTICAL FORECASTING OF EVENTMAGNITUDES

During stimulation, we applied in real time an event magni-tude forecasting model to guide operational decisions withrespect to induced seismicity. Hallo et al. (2014) introducedthe concept of seismic efficiency, SEFF, that describes the cor-relation between the cumulative moment release, ΣM0, and thecumulative injection volume ΔV :

EQ-TARGET;temp:intralink-;df1;323;409SEFF �ΣM0

μΔV; �1�

in which μ is the shear modulus, assumed to be 20 GPa here.Based on the observed values of SEFF and the b-value, the size ofthe largest expected event MMAX can be estimated as:

EQ-TARGET;temp:intralink-;df2;323;327MMAX�23

�log10

�SEFFμΔV �32−b�

b109:1

��23log10�10bδ−10−bδ�;

�2�in which δ is the probabilistic half-bin size defined aroundMMAX (Hallo et al., 2014). This formulation assumes that band SEFF do not change significantly for a given stage or for agiven volume of rock being stimulated. Verdon and Budge(2018) applied this approach in a pseudoprospective mannerto a hydraulic fracturing dataset from the Horn River Shale,Canada, showing that it would have accurately forecast eventmagnitudes had it been applied in real time.

Equation (2) posits a logarithmic dependence betweeninjection volume and the largest event size. Given that theplanned injection volumes do not vary by orders of magnitudebetween stages, the primary controlling factor on the largestevent magnitude is therefore SEFF. The relationship betweenSEFF, ΔV , andMMAX is plotted in Figure 7 (assuming b � 1).

Equation (2) provides the most likely maximum eventmagnitude. In practice, it is more useful to define a value

▴ Figure 4. Magnitude–frequency distribution for all eventsreported in real time (gray dots). The observed distribution followsthe Gutenberg–Richter (G-R) distribution with a � 1:9 and b � 1:3(red line). We use both the Wiemer and Wyss (2000; referred asW-W) formulation (green dashed line) and a Kolmogorov–Smirnov (K-S) test (purple dashed line) to assess the overallmagnitude of completeness.



forMMAX that is unlikely to be exceeded. Using synthetic eventdistributions, Verdon and Budge (2018) showed that adding avalue of 0.5 to equation (2) is sufficient to capture 95% of thevariance between true and reconstructed model populations. Inour analysis, we applied this correction such that our resultsprovided a value that, within reasonable levels of certainty, willnot be exceeded.

We tracked b and SEFF in real time during every stage,providing regularly updated forecasts of MMAX. We computedthe b-value using the Aki (1965) maximum-likelihood approach,finding the minimum completeness threshold using aKolmogorov–Smirnov test at a 10% acceptance level to assessthe quality of fit between the observed magnitude distributionand the G-R relationship (Clauset et al., 2009; Williams andLe Calvez, 2013), requiring a minimum of 50 events for a reliablemeasurement (although with >39,000 events in 17 stages, thenumber of events passed this threshold very quickly for each stage).

Figure 8 shows a selection of results for this analysis whenperformed on a stage-by-stage basis, that is, considering ΣM0and ΔV associated with each individual stage. We find thatfor most of the stages, this approach provided accurate bounds,with the observed events falling within the modeled value of

MMAX. However, this is not always the case,as can be seen for stages 32 and 38 inFigure 9, for example.

The reason for this discrepancy is obviouswhen considered in the light of the observationsand interpretations of the microseismicity pre-sented in the Microseismic Observations sec-tion: the NEF-1 runs obliquely to the welland was intersected by multiple stages. It istherefore not appropriate to consider each stageindependently because the seismicity was causedby repeated injection into the same feature.Instead, as the NEF-1 feature was identified,we adjusted our approach to include the effectsof repeated injection, treating all injection andseismicity from stage 18 onward cumulatively(Fig. 10a). The value of SEFF was observed tostabilize very quickly at a value of approximatelylog10 SEFF ≈ −2, which produces a forecastMMAX of 1.7. The largest observed event atPNR-1z had a magnitude of M � 1:5.

For completeness, we also considered thecumulative impacts of the full injection volumeand seismicity from all the injection stages(Fig. 10b). This represents the worst-case scenarioif all of the injected fluid was inducing events on asingle seismogenic feature. Initial values for SEFFare low (log10 SEFF ≈ −3), and b-values are high(b > 1:5), giving MMAX < 1. From stage 18onward, we observed the hydraulic fracturinginteract with the NEF-1, producing an increasein SEFF to (log10 SEFF ≈ −2) and a decrease inb to ∼1. This produces an increase in MMAXto MMAX ≈ 2.

DISCUSSION

Operational Decision-MakingThe observations presented previously were used by CRL toguide its operational decision making, especially during the lat-ter injection stages in December, after the period of injectionhiatus in November 2018.

During hydraulic fracturing, placement of the proppantcannot begin until fracture breakdown has occurred and frac-tures begin to propagate. This typically requires a minimum of∼80 m3 of fluid. The proppant concentration is then graduallyincreased as the injection continues, such that the majority ofproppant is placed at the end of the stage. If a stage is termi-nated midway through by a TLS red-light event, only a smallproportion of the proppant will have been placed even if sev-eral hundred m3 of fluid have been injected. In effect, the stagewill therefore have been wasted and the environmental wateruse and seismic risk unnecessarily increased.

At PNR-1z, the modeling described earlier showed thatevents larger than M � 2 were not expected on the NEF-1given the observed b-values and seismic efficiency, and the

▴ Figure 5. (a) Map view and (b) cross section of microseismic events detectedduring real-time monitoring at PNR-1z. Events are colored by the sleeve numberwith which they are associated. The PNR-1z well profile is shown by the black line.



planned injection program. This forecast wasreported to the OGA in November 2018,and it falls within the objectives of seismicitymitigation set out by the OGA (minimizing feltevents and avoiding damaging events).However, the NEF-1 could be expected to con-tinue producing M > 0:5 red-light events thatwould terminate injection, preventing theplacement of proppant. CRL therefore decidedthat further injection into the sleeves that inter-sect the NEF-1 would be wasted, and inDecember 2018, CRL restarted injection instages 37–41 at the heel of the well. Basedon the seismicity mapping described in theMicroseismic Observations section, it washoped that these stages would pass to the eastof the NEF-1, allowing these stages to be com-pleted without interruption. Based on the fore-casting described in the Statistical Forecasting ofEvent Magnitudes section, CRL was able to doso with confidence that if these stages did inter-sect NEF-1, the levels of seismicity would notexceed the objectives set by the OGA, andtherefore injection could be conducted safely.

In reality, some of these latter stages didintersect the NEF-1, triggering two furtherTLS events with M > 0:5. However, the eventmagnitudes remained within the levels that hadbeen forecast, as described in the StatisticalForecasting of Event Magnitudes section, andwithin the overall regulatory objective to min-imize the number of felt events.

Seismic Efficiency and Seismogenic IndexThe seismogenic index (SI; Shapiro et al., 2010)is another parameter that is commonly used todescribe the relationship between injected vol-ume and seismicity. Whereas the SEFF param-

eter we use here scales the injection volume to thecumulative seismic moment release, the SI scales the injectionvolume to the number of events larger than a given magnitude.Because many previous studies have provided estimates of SI, itis of interest to compute this parameter for the PNR-1z datasetto facilitate a comparison. Our results are shown alongside theSEFF results in Figure 10, and we also plot the MMAX forecaststhat result (at 5% probability of exceedance level) using themethod described by Shapiro et al. (2010). We note that, asfound by Verdon and Budge (2018), SI follows a similar trendto log10 SEFF, which is not surprising because the total momentrelease will depend on the number of events that occur. Wealso find that the MMAX values derived from the SI measure-ments are larger than those derived from the SEFF measure-ments, as also found by Verdon and Budge (2018).

Dinske and Shapiro (2013) catalog SI values for a range ofinjection sites, finding values ranging from −9 < SI < 1. Themaximum value of SI obtained here is SI � −1:8, which is

▴ Figure 6. Maps showing the observations used to identify seismogenic struc-tures. (a) All events withM > 0 (dots colored by sleeve number according to Fig. 5),the cumulative seismic moment (contours), and the focal mechanisms of the larg-est events. (b) A map of the events that occurred during the injection hiatus from 3November to 7 December. We combine the largest events and the injection hiatusevents to map a plane striking at 237° and dipping at 70° (black-outlined box).

▴ Figure 7. Relationship between SEFF, ΔV , and MMAX given byequation (2) (assuming b � 1), showing the logarithmic depend-ence of MMAX on ΔV .



similar to many of the geothermal projects described by Dinskeand Shapiro (2013) but significantly larger than those obtainedfor hydraulic fracturing sites at Cotton Valley (east Texas) andin the Barnett Shale (northeast Texas). However,the values obtained for PNR-1z are similar to values foundby Verdon and Budge (2018) for hydraulic fracturing in theHorn River Basin, British Columbia, Canada, where−4 < SI < −1, and toward the lower end of the range foundby Schultz et al. (2018) for hydraulic fracturing sites in Alberta,Canada, where −2:5 < SI < −0:5. The most notable past caseof injection-induced seismicity in the United Kingdom forwhich SI values are available is the Rosemanowes Hot DryRock geothermal site, for which Li et al. (2018) found

maximum values of SI � −3:4, significantlylower than the values found for PNR-1z.

Scaling between Volume and CumulativeMoment ReleaseThe underlying assumption implicit to equa-tion (1) is that the cumulative seismic momentscales linearly with the injection volume.However, recent studies (e.g., Galis et al., 2017;De Barros et al., 2019) proposed alternativescaling factors and in particular that

EQ-TARGET;temp:intralink-;df3;370;613ΣM0∝V32: �3�

This scaling by an exponent of 1.5 is alsoimplicit to the Shapiro et al. (2010) SI approachbecause the logarithm of the seismic momentscales with 1:5 ×Mw . Discussion continues asto the most appropriate value of the scalingexponent between ΣM0 and V (e.g., Chen et al.,2018; De Barros et al., 2019).

In Figure 11a, we track the evolution of thecumulative moment release with the cumulativeinjection volume and estimate a least-squares fit(in log–log space) to these data for a relation-ship having the form:

EQ-TARGET;temp:intralink-;df4;370;434ΣM0 � αV n: �4�Our results are shown in Figure 11a. For theoverall dataset, we find a best-fit value ofn � 1:6. However, it is apparent that the datamay not be best described by a single value.Based on our observations of which stages causedreactivation of the NEF-1, combined with appar-ent changes in slope of Figure 11a, we divide thedata into three periods: stages 1–14, before reac-tivation of the NEF-1; stages 18–38, while reac-tivation of the fault was taking place; and stages39–41, which appeared to miss the NEF-1 at theheel of the well. Doing so, we find best-fit valuesof n � 0:8 for stages 1–14, n � 3:0 for stages18–38, and n � 0:6 for stages 39–41.

This variability highlights a challenge that arises whenattempting to assess any scaling relationship between cumula-tive moment and volume if the constant of proportionality(α in equation 4) varies during the process, which might beexpected as hydraulic fracturing proceeds along a horizontalwell, encountering different volumes of rock that have differentgeomechanical properties.

We further demonstrate this effect in Figure 11b. Basedon our observations in the Statistical Forecasting of EventMagnitudes section, we simulate a scenario whereby event pop-ulations are generated with b � 1:2 and log10 SEFF � −2:6(assuming a linear relationship between V and ΣM0) forthe first 1600 m3 of injection (representing stages 1–14),

▴ Figure 8. Examples of SEFF, b, andMMAX tracked during injection on a stage-by-stage basis for stages (a) 02 and (b) 39. In the lower panels, we track SEFF (blue)and b (purple), and in the upper panels, we plot the resulting values ofMMAX (blackline) compared with observed events (circles colored by magnitude relative to theTLS thresholds).



log10 SEFF � −1:7 for the second 1500 m3 of injection (rep-resenting stages 18–38), and log10 SEFF � −2:7 for the final1100 m3 of injection (representing stages 39–41). Eventsare generated stochastically to meet these criteria and areassumed to occur at random times within each of the specifiedperiods. We generate 1000 such populations, and inFigure 11b, we plot the median value of ΣM0 as a functionof V and the boundaries containing 95% of the models.The resulting models show good agreement with the observedevolution of cumulative moment release.

This modeling indicates the need for caution whenattempting to constrain the relationship between momentrelease and volume: if the constant of proportionality varies

during injection, then a simple comparison ofmoment and volume may lead to under- or over-estimates of the exponent n. For this dataset, alinear relationship between cumulative momentand volume, with an increase in SEFF fromlog10 SEFF � −2:6 to log10 SEFF � −1:7 duringreactivation of the NEF-1, provides a good fit tothe observed seismicity.

Assigning Injection Volumes to SeismicityVerdon and Budge (2018) treated eachhydraulic fracturing stage as an independentevent and did not treat the volumes cumula-tively as injection stages proceeded. In contrast,for the PNR-1z dataset, Figures 9 and 10 showthe importance of treating multiple stages in acumulative manner and that failure to do sowould have produced a significant under-estimate of the expected event magnitudes forsome stages. We believe that the difference inbehaviors between the two sites stems fromthe orientations of the faults relative to the welltrajectories. In the Horn River Basin sitedescribed by Verdon and Budge (2018), thereactivated faults were orientated roughlyperpendicular to the wells. As such, each seis-mogenic feature was only affected by one ortwo stages (Kettlety et al., 2019). In contrast,for PNR-1z, the NEF-1 runs obliquely to thewell, so this feature was intersected by multiplefracture stages, hence the need to treat thesestages cumulatively.

Assigning the appropriate fluid volumewhen making such assessments remains achallenging issue (e.g., Atkinson et al., 2016).The comparison of the Horn River Basin andPNR-1z examples described previously showsthat detailed analysis of microseismic eventlocations, combined with a geomechanicalunderstanding of the subsurface, is needed toguide such decisions.

CONCLUSIONS

Recent hydraulic fracturing operations at the PNR-1z wellwere subject to some of the most stringent regulations regard-ing induced seismicity ever applied to any kind of industrialactivity. The operator therefore took a proactive approachto the issue, using real-time microseismic monitoring to makeoperational decisions with respect to induced seismicity.Microseismic observations allowed us to identify the presenceof a pre-existing structure on which elevated levels of seismicitywas occurring and to map its extent in the subsurface. Thisstructure produced multiple events that were above the TLSred-light threshold, forcing the operator to stop injection,resulting in wasted stages, when fluid injection ceased before

▴ Figure 9. Examples of SEFF, b, andMMAX tracked during injection on a stage-by-stage basis for stages (a) 32 and (b) 38, in the same format as Figure 8. For somestages, events occur that exceed the modeled MMAX values, when the injectionvolumes and observed events are treated discretely on a stage-by-stage basis.



significant quantities of proppant could be placed. Using themicroseismic observations, the operator was able to move toinjection locations that were less likely to interact with thisstructure, thereby increasing the chance of conducting success-ful stages.

At the same time, we used the microseismic observationsto populate a statistical model to estimate an upper bound for

the largest expected event size during injection.This model was successful in forecasting themagnitudes of the events that did occur. Theforecast maximum magnitudes of MMAX < 2was within the overall objective set by the regu-lator to minimize the number of felt events andeliminate the possibility of damaging events.This modeling gave the operator and the regu-lator confidence that even if the seismogenicstructure were to be intersected by further frac-turing stages, the level of risk posed was accept-able. This confidence was borne out duringoperations: further activity did occur on theidentified fault, but the largest event to occurhad a magnitude of M � 1:5, within theexpectations provided by the statistical model.

Various options have been suggested toregulate induced seismicity. Fault respect distan-ces (Westwood et al., 2017) require an operatorto avoid known faults in the subsurface.However, this case study, along with previouscases (e.g., Igonin et al., 2019; Kettlety et al.,2019), shows that reactivated faults may not bevisible on 3D seismic surveys, especially if theyhave strike-slip displacement. In contrast,imaged faults may not be near their criticalstress and therefore do not reactivate.Therefore, the use of fault-respect distances willnot provide an effective approach to inducedseismicity regulation.

Although more advanced approaches to themitigation of induced seismicity have been pro-posed (e.g., Mignan et al., 2017; Verdon andBudge, 2018) and demonstrated (Kwiatek et al.,2019), simple TLSs are the most common formof regulation applied by regulators to mitigateHF-IS. The retroactive nature of these TLSsmeans that red-light thresholds may be set farlower than the actual level of seismicity that aregulator wishes to prevent. Decisions are basedsolely on the magnitude of the largest events,which is a reasonable choice if sites are moni-tored by regional arrays that provide limiteddetection thresholds and poorly constrainedevent locations. However, where operatorsacquire high-quality real-time microseismicdata, providing thousands of accurately locatedevents across several orders of magnitude, then aTLS that use only the largest event magnitude,

and therefore discards 99.9% of the observations available,seems unnecessarily crude. In this article, we demonstratedhow an operator can use microseismicity to assess the seismicrisk and make proactive decisions to mitigate induced seismic-ity in real time. Such an approach is more in line with the typeof goal-setting regulation (Lindøe et al., 2012) that has beenapplied with much success to other aspects of the oil and

▴ Figure 10. Forecasting MMAX over cumulative stages. Here, we treat stagescumulatively to generate MMAX forecasts when (a) all of the stages that intersectthe northeast fault-1 (NEF-1) are considered and (b) when all stages are consid-ered. The observed SEFF is initially at approximately log10 SEFF ≈ −3, giving a fore-cast MMAX < 1. As the injection begins to interact with the NEF-1 feature, the b-value decreases, and the overall seismic efficiency increases to approximatelylog10 SEFF ≈ −2, giving a forecast MMAX < 2. (b) also shows the Shapiro et al.(2010) seismogenic index (SI, gold dashed line) and the resulting MMAX forecastfrom this approach (gray dashed line).



gas industry. Induced seismicity poses a risk for other forms ofsubsurface industrial activity including engineered geothermalsystems and the storage of CO2 in geologic reservoirs. Asinduced seismicity continues to attract public scrutiny, the pro-active real-time use of seismic monitoring, as demonstratedhere, could see many other applications.

DATA AND RESOURCES

The event catalogs and injection data used in thisarticle are available from the Oil and GasAuthority (https://www.ogauthority.co.uk/onshore/onshore‑reports‑and‑data/preston‑new‑road‑pnr‑1z‑hydraulic‑fracturing‑operations‑data/, last accessedJuly 2018).

ACKNOWLEDGMENTS

The authors thank Cuadrilla Resources and theirjoint venture partners Spirit Energy and A JLucas for collaboration on this project. Theauthors thank the British Geological Survey(BGS) for providing data from the additionalsurface stations that were installed at the site.The authors also thank Schlumberger forprocessing the monitoring data in real time.J. P. V. was funded by Natural EnvironmentResearch Council (NERC) Grant NumberNE/R018162/1, T. K. was supported by theNERC GW4+ Doctoral Training Partnership(Grant Number NE/L002434/1), and J.-M. Kand A. F. B. were funded by NERC GrantNumber NE/R018006/1.

REFERENCES

Aki, K. (1965). Maximum likelihood estimate of b inthe formula logN � a–bM and its confidencelimits, Bull. Earthq. Res. Inst. Univ. Tokyo 43,237–239.

Akkar, S., M. A. Sandikkaya, and J. J. Bommer (2014).Empirical ground-motion models for point- andextended-source crustal earthquake scenarios inEurope and the Middle East, Bull. Earthq. Eng.12, 359–387.

Atkinson, G. M., D. W. Eaton, H. Ghofrani, D. Walker,B. Cheadle, R. Schultz, R. Shcherbakov, K.Tiampo, J. Gu, R. M. Harrington, Y. Liu, M.van der Baan, and H. Kao (2016). Hydraulic frac-turing and seismicity in the Western CanadaSedimentary Basin, Seismol. Res. Lett. 87, 631–647.

Baisch, S., C. Koch, and A. Muntendam-Bos (2019).Traffic light systems: To what extent can inducedseismicity be controlled, Seismol. Res. Lett. 90,no. 3, 1145–1154.

Bao, X., and D. W. Eaton (2016). Fault activation byhydraulic fracturing in western Canada, Science354, 1406–1409.

Bosman, K., A. Baig, G. Viegas, and T. Urbancic(2016). Towards an improved understanding of

induced seismicity associated with hydraulic fracturing, FirstBreak 34, 61–66.

Broccardo, M., A. Mignan, S. Wiemer, B. Stojadinovic, and D. Giardini(2017). Hierarchical Bayesian modeling of fluid-induced seismicity,Geophys. Res. Lett. 44, 11,357–11,367.

Butcher, A., R. Luckett, J. P. Verdon, J-M. Kendall, B. Baptie, and J.Wookey (2017). Local magnitude discrepancies for near-event

▴ Figure 11. Evolution of cumulative seismic moment with injection volume. In (a),data points are colored by the corresponding stage. Power law fits to the obser-vations are shown, with a best-fit exponent of n � 1:61 for the overall dataset (darkgray dashed line) but n � 0:77 for the early stages, n � 2:98where the NEF-1 reac-tivates, and n � 0:58 during the final stages (orange dashed lines). In (b), the redcurve shows the observed data, and the black line shows the median stochasti-cally simulated model as described in Scaling between Volume and CumulativeMoment Release section, with the shaded region representing 95% of the models.



https://www.ogauthority.co.uk/onshore/onshore-reports-and-data/preston-new-road-pnr-1z-hydraulic-fracturing-operations-data/






receivers; implications for the UK traffic light scheme, Bull. Seismol.Soc. Am. 107, 532–541.

Chambers, K., J-M. Kendall, S. Brandsberg-Dahl, and J. Rueda (2010).Testing the ability of surface arrays to monitor microseismic activity,Geophys. Prospect. 58, 821–830.

Chen, X., J. Haffener, T. H. W. Goebel, X. Meng, Z. Peng, and J. C.Chang (2018). Temporal correlation between seismic moment andinjection volume for an induced earthquake sequence in centralOklahoma, J. Geophys. Res. 123, 3047–3064.

Clarke, H., L. Eisner, P. Styles, and P. Turner (2014). Felt seismicity asso-ciated with shale gas hydraulic fracturing: The first documentedexample in Europe, Geophys. Res. Lett. 41, 8308–8314.

Clauset, A., C. R. Shalizi, and M. E. J. Newman (2009). Power-law distri-butions in empirical data, Soc. Indust. Appl. Math. Rev. 51, 661–703.

Cuadrilla Resources Ltd. (2018). Preston New Road 1z Hydraulic FracturePlan, available at https://consult.environment‑agency.gov.uk/onshore‑oil‑and‑gas/information‑on‑cuadrillas‑preston‑new‑road‑site/supporting_documents/Preston%20New%20Road%20HFP.pdf (last accessed July2019).

De Barros, L., F. Cappa, Y. Guglielmi, L. Duboeuf, and J-R. Grasso(2019). Energy of injection-induced seismicity predicted from in-situ experiments, Nature Scientif. Rep. 9, 4999.

Dinske, C., and S. Shapiro (2013). Seismotectonic state of reservoirsinferred from magnitude distributions of fluid-induced seismicity,J. Seismol. 17, 13–25.

Edwards, B., and J. Douglas (2014). Magnitude scaling of induced earth-quakes, Geothermics 52, 132–139.

Eyre, T. S., D. W. Eaton, M. Zecevic, D. D'Amico, and D. Kolos (2019).Microseismicity reveals fault activation before Mw 4.1 hydraulic-fracturing induced earthquake, Geophys. J. Int. 218, 534–546.

Fellgett, M. W., A. Kingdon, J. D. O. Williams, and C. M. A. Gent(2017). State of stress across UK regions, Open-File Rept. OR/17/048, British Geological Society, Nottingham, United Kingdom.

Friberg, P. A., G.M. Besana-Ostman, and I. Dricker (2014). Characterisationof an earthquake sequence triggered by hydraulic fracturing in HarrisonCounty, Ohio, Seismol. Res. Lett. 85, 1295–1307.

Galis, M., J. P. Ampuero, P. M. Mai, and F. Cappa (2017). Induced seis-micity provides insight into why earthquake ruptures stop, Sci. Adv.3, eaap7528.

Green, C. A., P. Styles, and B. J. Baptie (2012). Preese Hall Shale GasFracturing Review and Recommendations for Induced SeismicMitigation, Department of Energy and Climate Change, London.

Grigoli, F., S. Cesca, A. P. Rinaldi, A. Manconi, J. A. López-Comino, J. F.Clinton, R. Westaway, C. Cauzzi, T. Dahm, and S. Wiemer (2018).The November 2017Mw 5.5 Pohang earthquake: A possible case ofinduced seismicity in South Korea, Science 360, 1003–1006.

Gutenberg, B., and C. F. Richter (1944). Frequency of earthquakes inCalifornia, Bull. Seismol. Soc. Am. 34, 185–188.

Hallo, M., I. Oprsal, L. Eisner, and M. Y. Ali (2014). Prediction of mag-nitude of the largest potentially induced seismic event, J. Seismol.18, 421–431.

Hammack, R.,W. Harbert, S. Sharma, B. Stewart, R. Capo, A. Wall, A.Wells, R. Diehl, D. Blaushild, J. Sams, and G. Veloski (2014). Anevaluation of fracture growth and gas/fluid migration as horizontalMarcellus shale gas wells are hydraulically fractured in GreeneCounty, Pennsylvania, EPAct Technical Report Series, NETL-TRS-3-2014, U.S. Department of Energy, National EnergyTechnology Laboratory, Pittsburgh, Pennsylvania.

Hansard (2018). House of Commons Official Report 649(208), 714–716.Häring, M. O., U. Schanz, F. Ladner, and B. C. Dyer (2008).

Characterisation of the Basel 1 enhanced geothermal system,Geothermics 37, 469–495.

Hofmann, H., G. Zimmermann, M. Farkas, E. Huenges, A. Zang, M.Leonhardt, G. Kwiatek, P. Martinez-Garzon, M. Bohnhoff, K-B.Min, et al. (2019). First field application of cyclic soft stimulationat the Pohang enhanced geothermal system site in Korea, Geophys. J.Int. 217, 926–949.

Hofmann, H., G. Zimmermann, A. Zang, and K.-B. Min (2018). Cyclicsoft stimulation (CSS): A new fluid injection protocol and trafficlight system to mitigate seismic risks of hydraulic stimulation treat-ments, Geotherm. Energy 6, 27.

Igonin, N., J. P. Verdon, J-M. Kendall, and D. W. Eaton (2019). The roleof parallel fracture networks for induced seismicity in the DuvernayFormation, EAGE Annual Meeting, London,United Kingdom, 3–7June 2019, Expanded Abstracts WS16_07.

Kendall, J-M., A. Butcher, A. L. Stork, J. P. Verdon, R. Luckett, and B. J.Baptie (2019). How big is a small earthquake? Challenges in deter-mining microseismic magnitudes, First Break 37, 51–56.

Keranen, K. M., M. Weingarten, G. A. Abers, B. A. Bekins, and S. Ge(2014). Sharp increase in central Oklahoma seismicity since 2008induced by massive wastewater injection, Science 345, 448–451.

Kettlety, T., J. P. Verdon, M. J. Werner, J-M. Kendall, and J. Budge(2019). Investigating the role of elastostatic stress transfer duringhydraulic fracturing-induced fault reactivation, Geophys. J. Int.217, 1200–1216.

Kwiatek, G., T. Saamo, T. Ader, F. Bluemle, M. Bohnhoff, M.Chendorain, G. Dresen, P. Heikkinen, I. Kukkonen, P. Leary, et al.(2019). Controlling fluid-induced seismicity during a 6.1-km-deepgeothermal stimulation in Finland, Sci. Adv. 5, eaav7224.

Li, X., I. Main, and A. Jupe (2018). Induced seismicity at theUK ‘hot dryrock’ test site for geothermal energy production, Geophys. J. Int.214, 331–344.

Lindøe, P. H., M. Baram, and J. Paterson (2012). Robust Offshore RiskRegulation—An assessment of US,UK and Norwegian approaches,European Safety and Reliability Conference, Helsinki, Finland, 25–29 June 2012.

Luckett, R., L. Ottemoller, A. Butcher, and B. Baptie (2019). Extendinglocal magnitude ML to short distances, Geophys. J. Int. 216, 1145–1156.

Maxwell, S. C., M. Jones, R. Parker, S. Miong, S. Leaney, D. Dorval, D.D'Amico, J. Logel, E. Anderson, and K. Hammermaster (2009).Fault activation during hydraulic fracturing, SEG AnnualMeeting Expanded Abstracts 28, 1552–1556.

Maxwell, S. C., J. Rutledge, R. Jones, and M. Fehler (2010). Petroleumreservoir characterization using downhole microseismic monitoring,Geophysics 75, A129–A137.

Maxwell, S. C., J. Shemeta, E. Campbell, and D. Quirk (2008).Microseismic deformation rate monitoring, SPE Annual TechnicalConference, Denver, Colorado, 21–24 September 2008, SPE116596.

Mignan, A., M. Broccardo, S. Wiemer, and D. Giardini (2017). Inducedseismicity closed-form traffic light system for actuarial decision-making during deep fluid injections,Nature Scientific Rep. 7, 13607.

Oil and Gas Authority (2018).Consolidated Onshore Guidance, Version 2.2,Oil and Gas Authority, London, available at https://www.ogauthority.co.uk/media/4959/29112017_consolidated‑onshore‑guidance‑compendium_vfinal‑002.pdf (last accessed April 2019).

Schultz, R., G. Atkinson, D. W. Eaton, Y. J. Gu, and H. Kao (2018).Hydraulic fracturing volume is associated with induced earthquakeproductivity in the Duvernay play, Science 359, 304–308.

Schultz, R., V. Stern, M. Novakovic, G. Atkinson, and Y. J. Gu (2015).Hydraulic fracturing and the Crooked Lake sequences: Insightsgleaned from regional seismic networks, Geophys. Res. Lett. 42,2750–2758.

Segall, P. (1989). Earthquakes triggered by fluid extraction, Geology 17,942–946.

Shapiro, S. A., C. Dinske, and C. Langenbruch (2010). Seismogenic indexand magnitude probability of earthquakes induced during reservoirfluid stimulations, TLE 29, 304–309.

Skoumal, R. J., M. R. Brudzinski, and B. S. Currie (2015). Induced earth-quakes during hydraulic fracturing in Poland Township, Ohio, Bull.Seismol. Soc. Am. 105, 189–197.

van der Elst, N. J., M. T. Page, D. A. Weiser, T. H. W. Goebel, and S.M.Hosseini (2016). Induced earthquake magnitudes are as large as (sta-tistically) expected, J. Geophys. Res. 121, 4575–4590.



https://consult.environment-agency.gov.uk/onshore-oil-and-gas/information-on-cuadrillas-preston-new-road-site/supporting_documents/Preston%20New%20Road%20HFP.pdf







https://www.ogauthority.co.uk/media/4959/29112017_consolidated-onshore-guidance-compendium_vfinal-002.pdf






Verdon, J. P. (2014). Significance for secure CO2 storage of earthquakesinduced by fluid injection, Environ. Res. Lett. 9, 064022, doi:10.1088/1748-9326/9/6/064022.

Verdon, J. P., and J. Budge (2018). Examining the capability of statisticalmodels to mitigate induced seismicity during hydraulic fracturing ofshale gas reservoirs, Bull. Seismol. Soc. Am. 108, 690–701.

Wang, Z., and A. Krupnick (2013). A retrospective review of shale gasdevelopment in the United States: What lead to the boom?,Resources for the Future DP 13-12 42 pp.

Webster, B. (2018). Cuadrilla to resume fracking seven years after trem-ors, The Times, 16 October 2018, available at https://www.thetimes.co.uk/article/cuadrilla‑to‑resume‑fracking‑seven‑years‑after‑tremors‑h6lrndhxj (last accessed March 2019).

Wessels, S. A., A. De La Peña, M. Kratz, S. Williams-Stroud, andT. Jbeili (2011). Identifying faults and fractures in unconventionalreservoirs through microseismic monitoring, First Break 29,99–104.

Westwood, R. F., S. M. Toon, P. Styles, and N. J. Cassidy (2017).Horizontal respect distance for hydraulic fracturing in the vicinityof existing faults in deep geological reservoirs: a review and mod-elling study, Geomechanics and Geophysics for Geo-Energy and Geo-Resources 3, 379–391.

Wiemer, S., and M. Wyss (2000). Minimum magnitude of completenessin earthquake catalogs: Examples from Alaska, the western UnitedStates, and Japan, Bull. Seismol. Soc. Am. 90, 859–869.

Williams, M. J., and J. Le Calvez (2013). Reconstructing frequency-mag-nitude statistics from detection limited microseismic data, Geophys.Prospect. 61, 20–38.

Zang, A., G. Zimmermann, H. Hofmann, O. Stephansson, K.-B. Min,and K. Y. Kim (2019). How to reduce fluid-injection-inducedseismicity, Rock Mech. Rock Eng. 52, 475–493.

Zinno, R., J. Gibson, R. N. Walker Jr., and R. J. Withers (1998).Overview: Cotton Valley hydraulic fracture imaging project,SEG Annual Meeting Expanded Abstracts 17, 338–341.

Huw ClarkeCuadrilla Resources Ltd.

Cuadrilla House, 6 Sceptre CourtBamber Bridge

Lancashire PR5 6AW, United [email protected]

James P. VerdonTom KettletyAlan F. Baird

J-Michael KendallSchool of Earth Sciences

University of BristolWills Memorial Building

Queen’s RoadBristol BS8 1RJ, United Kingdom

[email protected]@[email protected]

[email protected]

Published Online 31 July 2019



http://dx.doi.org/10.1088/1748-9326/9/6/064022

https://www.thetimes.co.uk/article/cuadrilla-to-resume-fracking-seven-years-after-tremors-h6lrndhxj





real-time imaging, forecasting, and management of human …gljpv/pdfs/clarke_etal_2019_srl.pdf ·...

Documents